Pith. sign in

REVIEW 4 major objections 5 minor 41 references

Disentangling Knowledge States with Ability and Proficiency Modeling for Knowledge Tracing

T0 review · 4 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

Pith's one-line read Splitting student histories into ability and proficiency phases improves knowledge tracing across six benchmarks.

desk verdict PAKT is a solid, modest KT contribution with a genuinely new phase-decomposition trick, but the causal story is hand-wavy, the masking isn't what it claims, and the lack of error bars leaves the headline gains unquantified. read the letter →

arxiv 2607.13103 v1 pith:GRJRLXW7 submitted 2026-07-14 cs.LG cs.AI

classification cs.LGcs.AI
keywords knowledgetracingabilityphaseproficiencydecompositionTransformercausalanalysisstudentmodelingintelligenttutoringsystems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a student's interaction history is not one uniform process: early attempts at a concept build ability, while later attempts reflect proficiency. It proposes to split each student's per-concept history at the point where they have given a threshold number of correct answers, then feed the two halves plus the original full sequence into separate branches of a Transformer. Across six public datasets, this phase-aware model outperforms all baselines on both AUC and accuracy, with gains up to 1.33%. A sympathetic reader would care because it suggests that explicit stage modeling, not just bigger or deeper sequence models, is what drives predictive gains in knowledge tracing.

What carries the argument

The load-bearing mechanism is the phase decomposition rule δ(s,c,t)=I(γ(s,c,t)<k), which assigns each interaction to the ability phase until the student has k correct answers on that concept, then to the proficiency phase. Paired mask matrices zero out the other phase's entries, producing ability, proficiency, and complete input branches for a two-stage Transformer (shared first decoder stage, branch-specific second stage). A type-aware readout then fuses the three branch representations with a learned mixing coefficient α and the residual difference H_F − H_A, which encodes practice-induced gain; a final MLP maps the concatenation to the response probability. The causal analysis claims this

What would settle it

Compare PAKT with a version whose phase labels are shuffled or replaced by random splits of the same length while keeping the identical three-branch architecture; if the gains over the full-sequence baseline persist for the shuffled labels, the phase semantics are not what drives performance. Also, on a dataset with a high early-correct-by-guessing rate, a k=1 split should hurt; if it does not, the transition assumption is not binding.

Watch

Extended reading notes

Core claim

The paper's central claim is that knowledge tracing improves when the model separately encodes the ability-building phase and the proficiency phase of each knowledge concept, rather than treating the whole response history as a single stream. The decomposition uses a threshold k: an interaction is assigned to the proficiency phase once the student has answered that concept correctly k times, otherwise to the ability phase. Three branches—ability, proficiency, and the complete original sequence—are encoded by a shared first-stage Transformer and then by branch-specific decoders; a readout module fuses them with a learned weight and an explicit proficiency-minus-ability difference vector. The

Load-bearing premise

The single load-bearing premise is that the point at which a student has answered a concept correctly k times marks the true boundary between ability-building and proficiency; if early correct answers are lucky, or the transition depends on recency or difficulty rather than raw correct count, the phase labels are noise and the model's gains would come from added capacity, not from phase modeling.

Editorial extensions

If this is right

  • If the phase decomposition is correct, any sequence-based KT model that ignores phase boundaries is mixing two causally different processes, so adding a phase split to an existing predictor should improve it without changing its core architecture.
  • The proficiency-minus-ability difference vector HΔ gives a per-concept, per-student quantity that could be used as a proxy for how much practice has consolidated mastery, enabling personalized practice recommendations.
  • The optimal k varies by dataset (1 to 4 in these experiments), meaning the transition point is learnable; a model that predicts the threshold per concept or per student should further improve over a fixed threshold.
  • Ablations show the complete branch is the most indispensable component, so phase modeling is complementary to, not a replacement for, the full history; hybrid architectures that layer phase-specific heads on a full-sequence backbone are a natural next step.
  • The maximum AUC gain is 1.33%, so even in the best case the improvement is modest but consistent, which is more notable for the principle than for raw effect size.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The threshold k is manually tuned and data-dependent; a testable extension is to replace it with a learned, per-concept transition predictor (e.g., based on recent correctness or spacing of practice) and check whether gains exceed the fixed-k version.
  • The causal analysis is conceptual—no intervention on ability or proficiency is measured; one could test the deconfounding claim by comparing PAKT against a version that randomizes the phase labels while keeping the same architecture, isolating whether gains come from the labels' semantics or just the extra branches.
  • The phase split uses only correct counts, ignoring question difficulty and response time; incorporating these could sharpen the transition boundary, especially on low-coverage datasets like Algebra05 where the current gain is smallest.
  • The sensitivity analysis predicts that a fully automatic phase model would show the smallest gains on high-concept-count, high-accuracy datasets such as Slepemapy, where the threshold choice barely matters—a pattern worth testing directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. PAKT proposes a knowledge tracing model that decomposes each student's per-concept interaction history into an ability phase and a proficiency phase based on the cumulative number of correct responses relative to a threshold k (Eq. 1-2). The decomposed masks are applied element-wise to input embeddings, and three branches—ability, proficiency, and complete—are processed by shared then branch-specific Transformer decoders. A type-aware readout module fuses the branches, including a proficiency-ability gap, before an MLP predicts response correctness. The paper reports consistent AUC/ACC improvements over six baselines on six datasets (maximum AUC gain 1.33%, average 0.82%), a sensitivity analysis for k, and ablation studies that remove individual branches. A causal analysis is presented to argue that separating ability and proficiency reduces confounding bias.

Significance. If the reported gains are robust, the phase-decomposition idea is a meaningful addition to KT: it goes beyond the coarse correctness-based split of DisKT and gives a behavioral interpretation to a multi-branch Transformer. The released code, five-fold cross-validation, multiple datasets, and ablation/sensitivity analyses are strengths. However, the improvements are small, no variance or significance measures are reported, and the central attribution of the gains to phase semantics is not yet supported by controlled experiments. The paper's value currently rests on an empirical pattern rather than on a validated causal mechanism for the phase decomposition.

major comments (4)
  1. [IV-A, Eq. (1)-(2); V-E; Fig. 7] The central claim that modeling ability vs. proficiency phases causes the reported gains is not established. The phase labels are defined by cumulative correct count—the same outcome variable the model predicts—and the threshold k is tuned per dataset (Section V-E). The ablations remove entire branches, changing model capacity, and no control experiment is run with random masks of identical sparsity, inverted masks, or masks based on alternative signals such as recency or difficulty. Without such controls, the improvements could come from extra model capacity or from the regularizing effect of zero-masking rather than from the phase semantics. The Conclusion explicitly concedes that k must be manually tuned and varies across datasets, which makes the attribution concern concrete. Please add controlled comparisons or substantially soften the causal attribution.
  2. [IV-B, Eq. (4)-(5); Eq. (6)-(7)] Element-wise zero-masking does not remove masked positions from Transformer attention. After zeroing the Q/K/V rows, the attention logits for those positions are zero rather than -inf, so they receive non-zero softmax weight and contribute as zero vectors; multi-head attention can also re-project them. Therefore the 'ability' and 'proficiency' branches are not strictly modeling only the designated subsequences. An explicit attention mask or sequence compaction is needed. This is not merely cosmetic: it affects the interpretation of the ablations and of 'separate modeling' throughout the paper.
  3. [Table II, V-D] Only mean AUC/ACC over five folds are reported, without standard deviations, confidence intervals, or significance tests. The largest gain is 1.33% AUC and the smallest 0.22%, while baseline-to-baseline differences are often of similar magnitude. Without variance or paired tests, the claim of consistent improvement is not statistically supported. In addition, the main results use dataset-specific k selected from the sensitivity analysis in Section V-E; unless this selection is nested inside the cross-validation, the reported gains may be optimistically biased. Please report per-fold variance and significance tests, and clarify how k is selected in the evaluation protocol.
  4. [IV-E, Eq. (15a)-(15e)] The causal derivation is not mathematically sound as written. The step from (15d) to (15e), 'focusing on the subspace d_s conditioned on specific ability a_s and proficiency f_s values,' replaces a marginalization/averaging operation without a well-defined event or expectation, and the notation conflates S as a random variable with S(a,f,d) as a deterministic function. The text calls (A,F)->D->S a 'backdoor path,' but in Fig. 4 D is a collider of A and F, so conditioning on D opens association rather than confounding. If the causal analysis is intended as a contribution, it should be rewritten rigorously; otherwise it should be removed or explicitly labeled as heuristic.
minor comments (5)
  1. [Figures 5-7] The figures appear as garbled glyph sequences rather than readable plots. If this reflects the actual submission, they must be replaced with legible figures; the prose descriptions alone are not sufficient.
  2. [Eq. (13)] The regularization term is underspecified: S^{A+}_u and S^{A-}_u are not precisely defined (how are positive/negative samples aggregated within a batch?), and no weight is given for L_reg in the total loss. Please clarify the computation and the relative weighting.
  3. [IV-A, Fig. 3] The figure label 'Professioncy Mask' appears to be a typo for 'Proficiency Mask.' Also, the statement 'Having multiple knowledge concepts does not impact the method's implementation' is not argued; since Eq. (1) is defined per concept c, the multi-concept case should be formally addressed or explicitly excluded from the problem formulation.
  4. [Fig. 1 caption] The caption says 'on more than a quarter of qualifying knowledge concepts' but does not define 'qualifying,' and the figure lacks axis labels. This makes the preliminary observation difficult to verify.
  5. [Fig. 4 caption] The caption refers to 'DAPKT' while the model is called PAKT throughout. Please correct the inconsistency.

Circularity Check

1 steps flagged · score 4.0 of 10

The reported benchmark gains are empirical and not circular, but the causal derivation in Eq. (15e) assumes the very ability/proficiency disentanglement it claims to establish.

  1. self definitional [Section IV-E, Eq. (15a)-(15e), especially the transition from Eq. (15d) to Eq. (15e)]
    "By focusing our analysis on the subspace d_s conditioned on specific ability a_s and proficiency f_s values, we arrive at the final estimable form in Eq.(15e)... P(a_s, f_s)P(Y|S(a_s, f_s, d_s), q). The probabilistic derivation in Eq.(15e) demonstrates that the student feature S is confounded by the backdoor path A, F → D → S."

    Eq. (15d) is a marginalization over all (a,f,d). The step to Eq. (15e) does not derive that sum; it simply selects one triple (a_s,f_s,d_s), asserts that the observed feature vector s corresponds to it, and then treats the resulting expression as the 'final estimable form.' That assertion is exactly the claim that ability and proficiency can be identified and separated from the observed history, the conclusion the causal analysis is supposed to prove. In the implemented model, A and F are not independently measured causes; they are masks built from cumulative correct counts (Eq. 1), so the causal claim reduces to a renamed version of the model's own input construction.

full rationale

PAKT's headline result—consistent AUC/ACC improvements over six baselines—is an external benchmark evaluation with 5-fold cross-validation; reported predictions on held-out folds are not forced by the phase definitions, and using historical correctness to construct features is standard supervised practice in KT, not circularity. The only genuine circular step I find is in the supporting causal analysis: the passage from Eq. (15d) to Eq. (15e) collapses the marginalization to a single (a_s,f_s,d_s) and thereby assumes the ability/proficiency disentanglement it claims to demonstrate. The paper itself concedes that the decomposition threshold k is manually tuned and dataset-dependent, which weakens the phase-semantics attribution but is a validity concern rather than circularity. The self-citation to SimpleKT (ref. [22], co-authored by Jinsong Chen) is used only for standard KC-sequence preprocessing and as a baseline, so it is not load-bearing. The missing random-mask control is an absence of evidence for the causal story, not a reduction of the prediction to its inputs. Overall score 4: central empirical claim retains independent content, but one supporting 'derivation' is circular by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the domain assumption that cumulative correct count is a valid phase boundary, on a single-concept simplification, on the causal graph being the true generative structure, and on the zero-masking operation preserving the semantics of separate subsequences. The free parameters k, alpha, truncation length, regularization weighting, and standard hyperparameters are tuned or learned, so the reported gains are partly the result of model selection.

free parameters (5)
  • k (phase threshold) = Dataset-specific optimum, not listed in main table; sensitivity indicates 1-5 depending on dataset
    The entire decomposition depends on the cumulative-correct-count threshold k (Eq. 1), tuned per dataset via validation.
  • alpha (fusion coefficient) = Learned scalar in [0,1]
    Eq. (8) balances ability and proficiency representations with a learnable coefficient, trained on the data.
  • Sequence truncation length = 200
    Section V-F states sequences are padded/truncated to length 200, which affects how much of the ability-to-proficiency transition is visible.
  • Regularization weight for L_reg = Not specified; implicitly 1
    The total loss in Eq. (14) sums L_BCE and L_reg with no explicit weighting, an unspecified modeling choice that affects training.
  • Standard hyperparameters (lr, dropout, embedding/hidden dims, layers) = Tuned per dataset on validation
    Section V-C reports hyperparameter search for every model-dataset pair; these are free parameters affecting the comparisons.
assumptions (5)
  • domain assumption Two-phase learning assumption: a student's interactions with a concept transition from ability-building to proficiency once they have accumulated k correct responses.
    Section IV-A, Eq. (1); the entire decomposition and branch design rests on this premise.
  • domain assumption Single-concept simplification: each question is assumed to contain one knowledge concept.
    Section IV-A states 'For simplicity, we consider only the scenario where each item contains one knowledge concept'; multi-concept datasets are handled by a KC-based conversion but the effect is not analyzed.
  • domain assumption Causal graph structure A,F -> D -> S -> Y and Q -> Y, with no other confounders.
    Section IV-E and Fig. 4; the graph is postulated, not learned or empirically validated.
  • standard math Standard probability manipulations: law of total probability, Bayes' rule, and S as a deterministic function of A, F, D.
    Section IV-E, Eq. (15b)-(15d); however the step to Eq. (15e) is not a valid instance of these rules.
  • ad hoc to paper Zero-masked rows behave as if the masked interactions are absent for Transformer attention.
    Eq. (4)-(5) element-wise multiply embeddings by binary masks but retain all positions; the paper calls these 'decomposed subsequences' without showing that attention ignores the zero rows.
invented entities (1)
  • Ability phase / Proficiency phase dichotomy
    purpose: Partition the student interaction history into two stages for separate modeling in the multi-branch Transformer.
    The two phases are defined purely by the cumulative-correct-count threshold k (Eq. 1). No external cognitive measurement or falsifiable consequence outside the paper validates that the split corresponds to distinct latent constructs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Disentangling Knowledge States with Ability and Proficiency Modeling for Knowledge Tracing." pith.science (2026). https://pith.science/paper/GRJRLXW7

@misc{pith2026260713103,
  author       = {Pith},
  title        = {Pith review of: Disentangling Knowledge States with Ability and Proficiency Modeling for Knowledge Tracing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GRJRLXW7}},
  note         = {Machine review of arXiv:2607.13103}
}
read the original abstract

Knowledge tracing (KT) aims to predict students' future performance by modeling their evolving knowledge states from historical interactions. Existing KT methods usually treat the raw interaction sequence as a unified behavioral process, overlooking the phase-specific nature of learning behaviors. Our preliminary observations show that students are more likely to correctly answer previously failed knowledge concepts after sufficient practice, suggesting a transition from ability-building to proficiency-oriented learning. Motivated by this, we propose Phase-Aware Knowledge Tracing (PAKT), a KT framework that decomposes student interactions into ability and proficiency phases based on the tailored decomposition mechanism. To effectively exploit the decomposed sequences, we design a multi-branch Transformer with a type-aware readout module to jointly capture phase-specific and holistic knowledge states. We further provide a causal analysis to reveal the confounding bias caused by entangling complex learning behaviors in phase-agnostic KT models. Extensive experiments on six public benchmarks demonstrate that our method consistently outperforms representative baselines, with a maximum AUC gain of 1.33% and an average gain of 0.82%.

Figures

Figures reproduced from arXiv: 2607.13103 by the authors.

Figure 1
Figure 1. Proportion of students exhibiting the two-phase learn [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. An overview of the PAKT model. The response correctness, question ID, and concept ID sequences are processed [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The illustration of the decomposition mask generation. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The causal graph of conventional KT model and [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Effect of parameter k on model performance . 0-50 50-100 100-150 150-200 Sequence Length Group 0.750 0.800 0.850 AUC better 0-50 50-100 100-150 150-200 Sequence Length Group 0.700 0.750 0.800 0.850 ACC better 0-50 50-100 100-150 150-200 Sequence Length Group 0.325 0.35…
Figure 6
Figure 6. Figure 6: Performance of PAKT Across Different Sequence Length Groups. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Ablation study on 6 datasets. However, PAKT has several limitations. First, when a stu￾dent practices too few interactions with a knowledge concept, the cumulative correct response count does not reach the threshold k, leaving little or no proficiency-phase signal for …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 1 linked inside Pith

  1. [1]

    A case study on the application of artificial intelligence in education industry,

    Q. Tang, “A case study on the application of artificial intelligence in education industry,” inProceedings of the 2023 3rd International Conference on Modern Educa- tional Technology and Social Sciences (ICMETSS 2023). Atlantis Press, 2023, pp. 99–109

  2. [2]

    Knowledge tracing: Modeling the acquisition of procedural knowledge,

    A. T. Corbett and J. R. Anderson, “Knowledge tracing: Modeling the acquisition of procedural knowledge,”User Modeling and User-Adapted Interaction, vol. 4, no. 4, pp. 253–278, 1994

  3. [3]

    F. M. Lord,Applications of Item Response Theory to Practical Testing Problems, 1st ed. New York: Rout- ledge, 1980

  4. [4]

    Deep knowledge tracing,

    C. Piech, J. Bassen, J. Huang, S. Ganguli, M. Sahami, L. J. Guibas, and J. Sohl-Dickstein, “Deep knowledge tracing,” inAdvances in Neural Information Processing Systems 28: Annual Conference on Neural Information Processing Systems 2015, December 7-12, 2015, Mon- treal, Quebec, Canada, C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett, Eds...

  5. [5]

    Deep-irt: Make deep learning based knowl- edge tracing explainable using item response theory,

    C. Yeung, “Deep-irt: Make deep learning based knowl- edge tracing explainable using item response theory,” inProceedings of the 12th International Conference on Educational Data Mining, EDM 2019, Montr ´eal, Canada, July 2-5, 2019, M. C. Desmarais, C. F. Lynch, A. Merceron, and R. Nkambou, Eds. International Educational Data Mining Society (IEDMS), 2019

  6. [6]

    A self attentive model for knowledge tracing,

    S. Pandey and G. Karypis, “A self attentive model for knowledge tracing,” inProceedings of the 12th Interna- tional Conference on Educational Data Mining, EDM 2019, Montr ´eal, Canada, July 2-5, 2019, M. C. Des- marais, C. F. Lynch, A. Merceron, and R. Nkambou, Eds. International Educational Data Mining Society (IEDMS), 2019

  7. [7]

    SINKT: A structure-aware inductive knowledge tracing model with large language model,

    L. Fu, H. Guan, K. Du, J. Lin, W. Xia, W. Zhang, R. Tang, Y . Wang, and Y . Yu, “SINKT: A structure-aware inductive knowledge tracing model with large language model,” inProceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21-25, 2024, E. Serra and F. Spezzano, Eds. ACM, 2024, pp....

  8. [8]

    Towards an appropriate query, key, and value computation for knowledge tracing,

    Y . Choi, Y . Lee, J. Cho, J. Baek, B. Kim, Y . Cha, D. Shin, C. Bae, and J. Heo, “Towards an appropriate query, key, and value computation for knowledge tracing,” in L@S’20: Seventh ACM Conference on Learning @ Scale, Virtual Event, USA, August 12-14, 2020, D. A. Joyner, R. F. Kizilcec, and S. Singer, Eds. ACM, 2020, pp. 341–344

Show all 41 references
  1. [9]

    I/O analysis is all you need: An I/O analysis for long- sequence attention,

    X. Lu, B. Long, X. Chen, Y . Han, and X. Sun, “I/O analysis is all you need: An I/O analysis for long- sequence attention,” inProceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2, ASPLOS 2026, P...

  2. [10]

    Dynamic key-value memory networks for knowledge tracing,

    J. Zhang, X. Shi, I. King, and D.-Y . Yeung, “Dynamic key-value memory networks for knowledge tracing,” in Proceedings of the 26th international conference on World Wide Web, 2017, pp. 765–774

  3. [11]

    Graph-based knowledge tracing: modeling student proficiency using graph neural network,

    H. Nakagawa, Y . Iwasawa, and Y . Matsuo, “Graph-based knowledge tracing: modeling student proficiency using graph neural network,” inIEEE/WIC/ACM international conference on web intelligence, 2019, pp. 156–163

  4. [12]

    Leveraging pedagogical theories to understand student learning pro- JOURNAL OF LATEX CLASS FILES, VOL. ??, NO. ??, ?? 202? 11 cess with graph-based reasonable knowledge tracing,

    J. Cui, H. Qian, B. Jiang, and W. Zhang, “Leveraging pedagogical theories to understand student learning pro- JOURNAL OF LATEX CLASS FILES, VOL. ??, NO. ??, ?? 202? 11 cess with graph-based reasonable knowledge tracing,” inProceedings of the 30th ACM SIGKDD Conference on Knowl...

  5. [13]

    EKT: exercise-aware knowledge tracing for student performance prediction,

    Q. Liu, Z. Huang, Y . Yin, E. Chen, H. Xiong, Y . Su, and G. Hu, “EKT: exercise-aware knowledge tracing for student performance prediction,”IEEE Trans. Knowl. Data Eng., vol. 33, no. 1, pp. 100–115, 2021

  6. [14]

    XKT: toward explainable knowledge trac- ing model with cognitive learning theories for questions of multiple knowledge concepts,

    C. Huang, Q. Huang, X. Huang, H. Wang, M. Li, K. Lin, and Y . Chang, “XKT: toward explainable knowledge trac- ing model with cognitive learning theories for questions of multiple knowledge concepts,”IEEE Trans. Knowl. Data Eng., vol. 36, no. 11, pp. 7308–7325, 2024

  7. [15]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997

  8. [16]

    Hyperbolic hypergraph transformer with knowledge state disentan- glement for knowledge tracing,

    J. Li, S. Mao, Y . Qin, F. Wang, and Y . Jiang, “Hyperbolic hypergraph transformer with knowledge state disentan- glement for knowledge tracing,”IEEE Trans. Knowl. Data Eng., vol. 37, no. 8, pp. 4677–4690, 2025

  9. [17]

    Stable knowledge tracing using causal inference,

    J. Zhu, X. Ma, and C. Huang, “Stable knowledge tracing using causal inference,”IEEE Trans. Learn. Technol., vol. 17, pp. 124–134, 2024

  10. [18]

    Cikt: Causality inspired knowledge tracing,

    S. Zu, L. Li, S. Cai, and J. Shen, “Cikt: Causality inspired knowledge tracing,” inInternational Confer- ence on Database Systems for Advanced Applications. Springer, 2024, pp. 485–495

  11. [19]

    Interpretable knowledge tracing with multiscale state representation,

    J. Sun, F. Yu, Q. Wan, Q. Li, S. Liu, and X. Shen, “Interpretable knowledge tracing with multiscale state representation,” inProceedings of the ACM on Web Conference 2024, WWW 2024, Singapore, May 13-17, 2024, T. Chua, C. Ngo, R. Kumar, H. W. Lauw, and R. K. Lee, Eds. ACM, 202...

  12. [20]

    Interpretable knowledge tracing via response influence- based counterfactual reasoning,

    J. Cui, M. Yu, B. Jiang, A. Zhou, J. Wang, and W. Zhang, “Interpretable knowledge tracing via response influence- based counterfactual reasoning,” in40th IEEE Interna- tional Conference on Data Engineering, ICDE 2024, Utrecht, The Netherlands, May 13-16, 2024. IEEE, 2024, pp. ...

  13. [21]

    Learning consistent representations with tem- poral and causal enhancement for knowledge tracing,

    C. Huang, H. Wei, Q. Huang, F. Jiang, Z. Han, and X. Huang, “Learning consistent representations with tem- poral and causal enhancement for knowledge tracing,” Expert Syst. Appl., vol. 245, p. 123128, 2024

  14. [22]

    sim- plekt: A simple but tough-to-beat baseline for knowledge tracing,

    Z. Liu, Q. Liu, J. Chen, S. Huang, and W. Luo, “sim- plekt: A simple but tough-to-beat baseline for knowledge tracing,” inThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net, 2023

  15. [23]

    Knowledge tracing: A survey,

    G. Abdelrahman, Q. Wang, and B. P. Nunes, “Knowledge tracing: A survey,”ACM Comput. Surv., vol. 55, no. 11, pp. 224:1–224:37, 2023

  16. [24]

    A survey of knowledge tracing: Models, variants, and applications,

    S. Shen, Q. Liu, Z. Huang, Y . Zheng, M. Yin, M. Wang, and E. Chen, “A survey of knowledge tracing: Models, variants, and applications,”IEEE Trans. Learn. Technol., vol. 17, pp. 1898–1919, 2024

  17. [25]

    Disen- tangled knowledge tracing for alleviating cognitive bias,

    Z. Yiyun, L. Zheqi, Z. Shengyu, and C. Jingyuan, “Disen- tangled knowledge tracing for alleviating cognitive bias,” inProceedings of the ACM on Web Conference, 2025, pp. 2633–2645

  18. [26]

    A three learning states bayesian knowledge tracing model,

    K. Zhang and Y . Yao, “A three learning states bayesian knowledge tracing model,”Knowl. Based Syst., vol. 148, pp. 189–201, 2018

  19. [27]

    Item response theory and clinical measurement,

    S. P. Reise and N. G. Waller, “Item response theory and clinical measurement,”Annual Review of Clinical Psychology, vol. 5, no. V olume 5, 2009, pp. 27–48, 2009

  20. [28]

    Context-aware attentive knowledge tracing,

    A. Ghosh, N. Heffernan, and A. S. Lan, “Context-aware attentive knowledge tracing,” inProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 2020, pp. 2330–2339

  21. [29]

    Con- trastive learning for knowledge tracing,

    W. Lee, J. Chun, Y . Lee, K. Park, and S. Park, “Con- trastive learning for knowledge tracing,” inProceedings of the ACM web conference 2022, 2022, pp. 2330–2338

  22. [30]

    Tracing knowledge instead of patterns: Stable knowledge tracing with diagnostic trans- former,

    Y . Yin, L. Dai, Z. Huang, S. Shen, F. Wang, Q. Liu, E. Chen, and X. Li, “Tracing knowledge instead of patterns: Stable knowledge tracing with diagnostic trans- former,” inProceedings of the ACM web conference 2023, 2023, pp. 855–864

  23. [31]

    Do we fully understand students’ knowl- edge states? identifying and mitigating answer bias in knowledge tracing,

    C. Cui, H. Ma, C. Zhang, C. Zhang, Y . Yao, M. Chen, and Y . Ma, “Do we fully understand students’ knowl- edge states? identifying and mitigating answer bias in knowledge tracing,”CoRR, vol. abs/2308.07779, 2023

  24. [32]

    Towards robust knowledge tracing models via k-sparse attention,

    S. Huang, Z. Liu, X. Zhao, W. Luo, and J. Weng, “Towards robust knowledge tracing models via k-sparse attention,” inProceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2023, Taipei, Taiwan, July 23-27, 2023, H....

  25. [33]

    P. M. Fitts and M. I. Posner,Human Performance. Belmont, CA, USA: Brooks/Cole Publishing Company, 1967, includes bibliographical references (pages 151- 158)

  26. [34]

    Disentangling user interest and conformity for recom- mendation with causal embedding,

    Y . Zheng, C. Gao, X. Li, X. He, Y . Li, and D. Jin, “Disentangling user interest and conformity for recom- mendation with causal embedding,” inProceedings of the web conference 2021, 2021, pp. 2980–2991

  27. [35]

    Deconfounded recommendation for alleviating bias am- plification,

    W. Wang, F. Feng, X. He, X. Wang, and T.-S. Chua, “Deconfounded recommendation for alleviating bias am- plification,” inProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, 2021, pp. 1717–1725

  28. [36]

    Algebra I 2005-2006 and Bridge to Algebra 2006-2007. Development data sets from KDD Cup 2010 Educational Data Mining Challenge,

    J. Stamper, A. Niculescu-Mizil, S. Ritter, G. Gordon, and K. R. Koedinger, “Algebra I 2005-2006 and Bridge to Algebra 2006-2007. Development data sets from KDD Cup 2010 Educational Data Mining Challenge,” 2010

  29. [37]

    Ad- dressing the assessment challenge with an online system that tutors as it assesses,

    M. Feng, N. T. Heffernan, and K. R. Koedinger, “Ad- dressing the assessment challenge with an online system that tutors as it assesses,”User Model. User Adapt. Interact., vol. 19, no. 3, pp. 243–266, 2009

  30. [38]

    Ednet: A large-scale hier- archical dataset in education,

    Y . Choi, Y . Lee, D. Shin, J. Cho, S. Park, S. Lee, J. Baek, C. Bae, B. Kim, and J. Heo, “Ednet: A large-scale hier- archical dataset in education,” inArtificial Intelligence in Education - 21st International Conference, AIED 2020, Ifrane, Morocco, July 6-10, 2020, Proceeding...

  31. [39]

    Adaptive geography practice data set,

    J. Papousek, R. Pel ´anek, and V . Stanislav, “Adaptive geography practice data set,”J. Learn. Anal., vol. 3, no. 2, pp. 317–321, 2016

  32. [40]

    Automatic discovery of cognitive skills to improve the prediction of student learning,

    R. V . Lindsey, M. Khajah, and M. C. Mozer, “Automatic discovery of cognitive skills to improve the prediction of student learning,” inAdvances in Neural Information Processing Systems 27: Annual Conference on Neu- ral Information Processing Systems 2014, December 8- 13 2014, ...

  33. [41]

    Dropout: a simple way to prevent neural networks from overfitting,

    N. Srivastava, G. E. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, “Dropout: a simple way to prevent neural networks from overfitting,”J. Mach. Learn. Res., vol. 15, no. 1, pp. 1929–1958, 2014

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.